d3ro-voice/scripts/verification/typing-fixture/README.md

15 lines
2.2 KiB
Markdown

# Windows typing fixture
Build without downloading dependencies:
```powershell
pwsh -File scripts/verification/typing-fixture/build.ps1
```
Open `test-results/typing-load/fixture/D3ROTypingFixture.exe` from the interactive desktop, then click **Start verification**. The default report is `test-results/typing-load/current-fixture.json`; an optional first executable argument chooses another report file.
The fixture only edits its own multiline textbox. It selects an already installed Korean input language on its own UI thread and restores that selection afterward. It never installs an IME, changes the default input method, reads other app documents, launches D3RO, or modifies D3RO configuration. Foreground/focus changes cancel injection; pressing Stop or closing the fixture cancels CPU workers. Each CPU phase stops after at most 20 seconds even if the UI is delayed.
Calibration must produce `안녕하세요 ` from the scan-code sequence `dkssudgktpdy ` and receive native `WM_IME_COMPOSITION` messages before any result is accepted. This prevents automatic Latin-to-Korean correction alone from masquerading as IME proof. The matrix uses 60/30/15 ms requested per-key spacing, both idle and optional CPU load. Read the recorded timestamps for achieved pacing. CPU load uses logical processor count minus one dedicated workers and records actual system/fixture CPU utilization. The sequence contains `안녕하세요 감사합니다 반갑습니다 ` repeated three times. The **Latin layout correction** option keeps the same keys and expected Korean text but switches this fixture's IME off after calibration; use this only when the D3RO configuration under test should automatically correct layout mistakes.
Record the application build/configuration under test in the label. Preserve each report under a distinct name before rerunning. Exact text match, native accepted-event counts, observed key counts, per-case text changes, focus validity, pacing and CPU overlap are separate evidence. Synthetic scan codes exercise OS input/IME and the global hook; they are not physical-hardware keyboard proof. This WinForms surface also does not prove Chromium/Electron or terminal behavior.